borghei/Claude-Skills / engineering/codex-cli-specialist
codex-cli-specialist
UnverifiedConverts Claude Code skills to OpenAI Codex format (agents/openai.yaml), checks dual-platform compatibility and builds a skills index. Use when porting or syncing skills to Codex, or setting up Codex CLI for a skill library.
Codex CLI Specialist
The agent converts Claude Code skills to Codex-compatible format, validates cross-platform compatibility, and builds skill registry manifests. It generates agents/openai.yaml configurations from SKILL.md frontmatter, runs up to 21 compatibility checks across both platforms, and produces skills-index.json for discovery systems.
Core Capabilities
- Skill conversion — turn a Claude Code SKILL.md into a Codex-compatible skill by generating
agents/openai.yamland copying scripts/references/assets. - Cross-platform validation — run up to 21 checks across Claude Code, Codex CLI, and shared categories (frontmatter, openai.yaml, encoding, naming, size).
- Index building — scan skill directories and emit a
skills-index.jsonmanifest for registries, discovery, and version pinning. - Codex CLI setup & operation — install, configure API access, and run with the right approval mode (suggest / auto-edit / full-auto).
- Dual-target authoring — write skills that work on both platforms from one shared
scripts/,references/,assets/tree. - Distribution & sync — keep Claude Code and Codex in sync via shared repo, CI/CD conversion, or git hooks; publish via GitHub releases.
When to Use
- Setting up or operating OpenAI Codex CLI.
- Converting or syncing skills between Claude Code and Codex.
- Configuring or reviewing
agents/openai.yaml. - Validating a skill for dual-platform compatibility.
- Building a
skills-index.jsonfor a skill library.
Clarify First
Before converting or building, settle the inputs below. When one is missing or vague and would change the result, ask about it; when it wouldn't, use a sensible default and say which one you used.
- Task: convert a skill, validate cross-platform compatibility, or build a skills index (selects
codex_skill_converter.pyvscross_platform_validator.pyvsskills_index_builder.py). - Source path: the SKILL.md or skill directory to operate on (the input the scripts read).
- Strictness: strict validation mode and whether to fail on warnings (sets
--strictand the pass/fail gate).
Ask about the two or three inputs that matter most for this request, not the whole list. If the user wants a result straight away, produce it and state your assumptions at the top.
Quick Start Workflow
# 1. Convert a skill. Without --output-dir it writes agents/openai.yaml into the source skill.
python scripts/codex_skill_converter.py ../code-tour/SKILL.md --output-dir /tmp/code-tour-codex
# 2. Validate both sides: Claude frontmatter, openai.yaml structure, encoding, naming, size
python scripts/cross_platform_validator.py /tmp/code-tour-codex # 21/21, exit 0
python scripts/cross_platform_validator.py ../code-tour # no openai.yaml yet, exit 1
# 3. Build a skills-index.json for a directory of skills
python scripts/skills_index_builder.py .. --output /tmp/skills-index.json --format human
Tools Overview
| Tool | Purpose | Key flags |
|---|---|---|
codex_skill_converter.py | Generate agents/openai.yaml from SKILL.md frontmatter and copy scripts/references/assets | --output-dir DIR, --json |
cross_platform_validator.py | Up to 21 checks (13 without agents/openai.yaml) across Claude Code, Codex and shared rules | --strict (warnings fail), --json |
skills_index_builder.py | Scan skill folders into a skills-index.json manifest | --output FILE, --format json|human, --category CAT |
The generated agents/openai.yaml leaves model unset so the skill inherits the user's Codex CLI
default; pin a model only when the skill truly needs one, using an id from OpenAI's current list.
Codex CLI flags and config file names have changed between releases, so confirm commands from
references/codex-cli-guide.md with codex --help on the installed version.
Exit codes. 0 converted / compatible / index written; 1 conversion failed or the skill is
not compatible (with --strict, warnings count); 2 the SKILL.md or directory does not exist.
Done when: the validator passes on the converted skill, a Codex session picks the skill up and runs one of its scripts, and the Claude Code copy still passes its own lint.
Anti-Patterns
Forking the skill per platform
Mistake: Keeping a Claude copy and a Codex copy of the same skill and editing them separately.
Why it happens: Converting once feels simpler than wiring a sync.
Instead: Keep one scripts/, references/ and assets/ tree and regenerate agents/openai.yaml from SKILL.md in CI or a hook.
Pinning model ids in skill config
Mistake: Hard-coding a model name in every openai.yaml.
Why it happens: Templates often ship with one filled in.
Instead: Omit model and inherit the CLI default; pinned ids go stale and break when a model is retired.
Trusting the conversion blindly
Mistake: Shipping the generated openai.yaml without reading it.
Why it happens: The converter prints "Conversion complete".
Instead: Read the instructions and tool list it extracted, then run the validator and one real Codex invocation.
Files
scripts/codex_skill_converter.py,scripts/cross_platform_validator.py,scripts/skills_index_builder.py— the three CLIs above.assets/openai-yaml-template.yaml— annotatedagents/openai.yamltemplate.references/— workflows, configuration and distribution, Codex CLI guide, cross-platform authoring, best practices.
Reference Documentation
Load the reference that matches the task — keep this file lean and pull detail on demand:
- references/workflows.md — Quick Start, Tools Overview, the 5 core workflows (install, author, convert, validate, build index), Common Patterns, and deep per-tool parameter/output reference. Read when running a workflow or needing exact tool arguments.
- references/configuration-and-distribution.md —
agents/openai.yamlstructure, discovery/locations, invocation patterns, cross-platform patterns, frontmatter compatibility, install/versioning, and sync/CI/CD/GitHub distribution. Read when configuring openai.yaml or distributing a library. - references/best-practices-and-troubleshooting.md — best practices (authoring, Codex usage, cross-platform, performance), anti-patterns, troubleshooting table, and success criteria. Read before shipping or when a tool misbehaves.
- references/codex-cli-guide.md — full Codex CLI reference: installation, configuration, approval modes, skill system, invocation, built-in features, environment variables.
- references/cross-platform-skills.md — writing skills for multiple agents (Claude Code, Codex, Cursor, Copilot, Goose): platform comparison, universal structure, per-platform config files, template.
- assets/openai-yaml-template.yaml — annotated Codex config template (instructions, tools, optional model, versioning).
Scope & Limitations
This skill covers:
- Installing, configuring, and operating OpenAI Codex CLI
- Converting Claude Code SKILL.md files into Codex-compatible format with
agents/openai.yaml - Validating skill directories for dual-platform (Claude Code + Codex CLI) compatibility
- Building skill registry manifests (
skills-index.json) for discovery and distribution
This skill does NOT cover:
- Writing the actual domain logic inside Python tool scripts (see senior-fullstack, code-reviewer, or the relevant domain skill)
- Cursor, Windsurf, Cline, or Aider platform-specific configuration (see standards/ and root-level dotfiles like
.cursorrules,.windsurfrules) - OpenAI API key management, billing, or rate-limit troubleshooting (out of scope -- refer to OpenAI documentation)
- Automated testing or CI/CD pipeline authoring beyond skill validation (see senior-devops and templates/)
See Also
- code-reviewer — a multi-script skill that makes a good first conversion test.
- claude-code-mastery — authoring the Claude Code side of a skill.
- senior-devops — wiring validation and index builds into CI.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 465k | Updated | Claude CodeCodex | Verified | Install |
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 465k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 318k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 318k | Updated | Claude CodeCodex | Verified | Install |